It’s an article of faith that technological innovation is crucial to prosperity and is currently changing our lives at an unprecedented rate, but how do we know if the pace of pioneering breakthroughs is any faster today than it was during Thomas Edison’s era? In fact, some economists argue that today’s information revolution has had much less impact on our lives than the big inventions of the late 19th century had on people living then.
The problem is that it’s very difficult to isolate the truly epic inventions from those that are incremental or trivial and thus to compare historic innovation trends. That in turn makes it hard to identify the policies or conditions that are likely to spur more breakthroughs.
Now, a team of researchers that includes SIEPR Senior Fellow and Stanford Graduate School of Business Professor Amit Seru has developed a novel strategy that applies big data computing to several million patent text documents to rank the innovative importance of almost every U.S. patent over the past 200 years — and to identify historical spikes of epic inventing.
For Silicon Valley fans, the good news is that one of those spikes has indeed been during the past two decades, and it’s been dominated by electronics and communications.
Elevators, Sewing Machines, and Combustion Engines
But the biggest surge, at least as measured by this new yardstick, came in the early to mid-1800s and featured inventions that revolutionized transportation, manufacturing, and the nature of big cities. Among the most important: vulcanized rubber, invented by Charles Goodyear, which became crucial in tires but also improved a vast range of industrial products; the elevator, invented by Elisha Otis, which made it practical to build skyscrapers and set the stage for modern cities; and the sewing machine, invented by Elias Howe, which transformed the garment industry.
A second big wave began in the late 1800s. The big breakthroughs in that era included the telephone in 1876; the internal combustion engine in 1877; the incandescent light bulb in 1880; the mechanical calculator and the first electric motor to run on alternating current, both in 1888. Orville Wright’s patent for the airplane, a blockbuster as measured in this study, came in 1906.
The real importance of the new paper, says Seru, is that it provides a powerful new metric to identify trends in innovation, which in turn opens doors toward understanding economic forces and policies that might foster more of it.
“Economists agree on the importance of technological progress when it comes to fostering economic activity, but we don’t really have many good ways of measuring it, especially over a long horizon,” Seru says. “It’s important to have robust ways of measuring technological innovation to understand how large changes in innovative activity move with policy changes. Is the right amount of innovation taking place? Is it shifting from large firms to small ones or vice versa? Is it coming from public firms or private firms? Is there a shift toward or away from universities or government agencies? These are all first-order questions and our metric opens avenues to study them.”
Searching Patents for Recurring Terms
The basic idea behind the new approach is fairly simple: An important invention is one that both differs greatly from what came before and greatly influences what comes later. To find those kinds of inventions, the researchers looked for patents with terms and phrases that appeared rarely in previous patents but showed up frequently in subsequent ones.
The computing challenge was immense. The researchers had to identify important terms in 9 million patents, each of which contains thousands of words. Then they had to analyze how frequently those terms showed up in each of the patents during previous and subsequent years. This large “correlation matrix” is what made the task computationally intensive. In the end, the ratio between those two measures became the gauge of a patent’s importance.
Why go to all that trouble?
Seru says the new approach has several advantages over existing strategies to measure innovation. The most popular of those strategies has been to count the number of times a patent is cited as “prior art” in patents that come later. The problem with that process is that patents didn’t consistently include such citations until the 1940s, which limits their usefulness in analyzing longer historical trends and answering those “first-order” questions.
A more recent approach, which Seru himself developed, is based on calculating the importance of an innovation based on the jump in a company’s stock price when it gets a new patent. His research shows that this metric of technological progress is a good indicator of real-world value, but with one significant caveat: It assesses the value of innovation only for publicly traded firms, since there are no stock prices available for an individual, a privately held company, a university, or a government agency.
Aligned with Historians
As it happened, the researchers found that their measures of patent quality based on textual analysis correlated quite well with the other measures — strong evidence that their phrase-based approach is reliable.
They also found that their top-ranked patents synced up well with the assessments of historians. Looking at a much-cited list of the 110 most important American patents up through the early 1960s, the researchers found that 40% ranked in the top 10% by their own measure of importance.
Many of the most legendary American inventions ranked in the top 1% of the new measure. These included some obvious ones, such as Thomas Edison’s electric light, Alexander Graham Bell’s telephone, and Wright’s airplane. But they also included less-remembered breakthroughs, such as the safety pin and Gail Borden’s invention of condensed milk.
Just slightly lower down the list, but still within the top 10%: Philo T. Farnsworth’s patent for the television, Edwin Armstrong’s FM radio, and Robert Noyce’s semiconductor.
The researchers also found evidence that technology innovation spurs economic productivity in a way that makes quantitative sense. That may sound obvious, but previous studies found it difficult to predict productivity using different metrics of technological innovation. The reason, says Seru, is that, unlike earlier studies, the textual based metric potentially enables researchers to filter out a cleaner measure of technological innovation — especially breakthrough innovations.
The big news, says Seru, is that the new way of ranking important inventions appears to be accurate and sets the stage for deeper understanding of the conditions in which real innovation can thrive.
This article first appeared in Insights by Stanford Business.